Automated volumetric model construction and dynamic segmentation of the heart ventricles in tagged MRI

Albert Montillo, University of Pennsylvania

Abstract

Cardiovascular disease is the leading cause of death for men and women in most developed countries, however tagged MRI (tMRI) may facilitate patient diagnosis and treatment planning and thereby lower morbidity. The imaging technique provides in vivo measurement of regional heterogeneity of the myocardium's contraction when the epicardial and endocardial surfaces and tag sheets are segmented. Since the adoption of tMRI has previously been hindered by lengthy analysis time for the epicardial and endocardial surfaces, we have developed a fully automated system which constructs a biventricular model of the heart and dynamically segments the ventricles throughout systole. After acquiring tMRI of our volunteers, we correct surface coil-induced intensity inhomogeneity using a scale based approach. We improve noise suppression by combining inhomogeneity correction and anisotropic diffusion. To extract 2D motion in each image, we apply an approach (called GAMA) based on banks of Gabor filters. We then extract ventricle surface features using statically adaptive, gray scale morphology. From the features, we construct a biventricular patient-specific model of the heart directly from the image data, which provides internal regularization forces. Then we derive 3D model forces from the 2D features and fit the model by balancing forces. We have found excellent agreement between our fitted models and expert delineations and our system replaces the 4 to 17 hours of interactive segmentation with just 1.75 hours of CPU time and it simultaneously fits a biventricular model throughout systole. Our MRI noise suppression is significantly better than previous methods and GAMA recovers all tag information including tag orientation and spacing. Our deformable model approach yields constraints between those of a model-from-data approach and parameter function/statistical modeling approaches and is suitable for processing steady state free precession images and 3D images. Our methods may be used for clinical applications including ventricular hypertrophy, myocardial infarcts, breathing disorders and speech impediments, or used for research applications including brain trauma, genetically engineered mouse hearts, blood flow analysis, and geometric statistical model construction.

Subject Area

Biomedical research|Computer science|Radiology

Recommended Citation

Montillo, Albert, "Automated volumetric model construction and dynamic segmentation of the heart ventricles in tagged MRI" (2004). Dissertations available from ProQuest. AAI3138055.
https://repository.upenn.edu/dissertations/AAI3138055

Share

COinS